To select an infrastructure-as-a-service environment, you first need to know what IaaS features you want from a public cloud provider.
While there are many IaaS offerings, the “Big Three” public clouds — Amazon Web Services (AWS), Microsoft Azure and Google Cloud Platform — together with IBM account for about 63% of market revenue, per a Synergy study published in February 2017.
A sizable number of other players, however, have emerged. These range from companies like Rackspace to smaller or more niche cloud vendors, such as DigitalOcean.
It can be complex to choose the right IaaS offering for your needs. This is partially because neither performance nor cost is standardized across the provider spectrum. Solid benchmarking work is still hard to find in the public cloud space, though companies like Cloud Spectator have ranking reports for the major U.S. and European providers, which compares them based on VM and storage performance.
Still, you must carefully evaluate and compare public IaaS vendors based on a number of IaaS features and factors. These include reliability, geographic scale, price and ability to support hybrid clouds.
VM options and performance
Public cloud providers’ VMs vary in performance and value, even when the configurations sound the same. Reasons for this include the use of different hardware, including systems with faster or more featured CPUs, as well as software and configuration differences that pass through into the VMs.
According to Cloud Spectator, raw VM performance can vary as much two times for the “same” configuration. The range is much bigger — up to eight times — for I/O performance, though this number is a single point for Rackspace, with all the others within a 3x range.
Google comes in at about 67% of the best performance of all the cloud service providers (CSPs) tested by Cloud Spectator, while AWS is at 75%. Azure is actually very close to the best value. When block disk is added to the performance, though, AWS is a poor 15% of Rackspace, the best performer, with Azure and IBM just a little better. Google came in at 45% of Rackspace.
From an IaaS features perspective, AWS is the clear leader with a rich ecosystem. Google and Microsoft are making strides to keep up. IBM has also built a strong portfolio, but in SaaS services more than basic tools for IaaS to build upon. The Big Three all offer a broad set of instance types and sizes, which they optimize variously for compute-heavy, memory-intensive or IO-driven apps. They also provide GPU instances and soon AI and machine learning acceleration with field-programmable gate arrays (FPGAs), matching much of the internal configurations that are used for these workloads. This is a key differentiation from smaller or niche-based CSPs, reflecting the fast deployment capabilities of the Big Three.
The most important takeaway from these numbers is that there is variation in the value of instances. The smartest approach is to sandbox your apps to get a personalized performance picture.
Reliability and geographic distribution
Even the Big Three IaaS providers suffer outages. Amazon’s Simple Storage Service (S3) had an outage in a North Virginia data center that lasted two days, but only for tenants that didn’t use multizone storage replication. Most outages at major providers are due to internal admin errors, such as failed router code updates.
If users distribute workloads across multiple provider availability zones, as well as replicate storage, they can mitigate most, if not all, of these outages. That said, the Big Three likely have the edge in availability over smaller CSPs, because they generally have better tested and more robust orchestration code, as well as a bigger pool of resources to help with recovery. In addition, major CSPs tend to have bigger geographic footprints in terms of data center locations, which increases user options around availability — although, even small CSPs usually have at least two data centers. Backup and archiving systems also benefit from a provider’s geodiversity.
However, some CSPs, including AWS, Google and Azure, charge extra for moving data and instances between zones and regions, so be sure to factor that into your total cost of ownership model.
Lastly, when assessing IaaS providers’ availability zones, you should consider their locality to markets. When you locate data in close proximity to consumers of that data, it reduces latency and cost. This is a major issue for multinational web presence.
Management and ease of use
From management dashboards to tools to create and tear down scripts, the ease of use of an IaaS platform affects the daily life of the admin. AWS, Azure and Google are heavily focused on add-on services for their clouds. While this increases operational options, it sometimes does so at the cost of simplicity. DigitalOcean, on the other hand, focuses more on ease of use for its core cloud offering, rather than additional services.
AWS, Azure, Google and IBM have all invested in tools and services to do a spectrum of jobs. This is an area to evaluate if you plan to integrate homegrown code and service packets, such as databases. More and more, the large CSPs support third-party SaaS apps to increase the potential value of these integrations. In this area, the largest CSPs will likely come out ahead.
These large CSPs also have well-developed tools for scripting and virtual networking, with AWS having a leadership role in the market. With AWS having much more compute capacity for IaaS than any competitor, it tends to drive the standard for tools and compatibility efforts in areas like scripting. An example is the interface to object storage. Every other major CSP has an S3 compatibility mode.
CSPs also offer management control via command-line interfaces or dashboards. To evaluate these management tools, set up a small sandbox to get a sense of each CSP’s capabilities. Then, build and manage a set of instances, both creating and killing them. Continue by moving data around to different storage systems. You should make the effort to benchmark performance running your apps. Look at costs and ease of use. For large-instance clusters, you’ll need to evaluate cluster management tools as well.
Some use cases require more testing. As the Cloud Spectator testing shows, compute performance is only a part of the story, with wide variations in I/O performance. If you have I/O-intensive workloads, you must measure storage. Some advanced workloads will need third-party tools. Examples are GPU- and FPGA-based instance clusters, where network performance and special computing may differ considerably depending on the implementation of underlying hardware.
Despite efforts around convergence, these tool sets aren’t fully interoperable yet, as APIs differ between vendors.
Many CSPs offer a range of data services, which might color your initial IaaS decision. For now, it seems the Big Three, plus IBM, are ahead of other smaller or more niche CSPs in terms of services to improve cloud productivity. Examples of these services include a variety of databases, load balancers, big data analytics and GPU support. AWS leads the pack, with Azure on its heels, while Google and IBM are aiming for the high-growth performance and intelligence market.
The same is true of GPU and AI instances, with IBM joining the Big Three with a range of offerings. This area of emerging technology demands high development costs, which could lock out the smaller CSPs.
Hybrid cloud support
The ability to support hybrid cloud is one of the more increasingly important IaaS features. New partnerships and services in this area might reinforce the power of market leaders like AWS and Azure and influence your choice of a CSP.
Microsoft launched Azure Stack to enable organizations to more seamlessly deploy Azure services across private and public clouds, while the partnership between AWS and VMware around VMware Cloud on AWS lets users deploy a single vSphere-managed hybrid cloud.
Google also now partners with Red Hat to enable hybrid cloud deployment.
The likely result of these relationships is that, within a couple of years, AWS, Azure and Google will have a role in hybrid cloud environments that users can manage from a single domain. The cost of developing software for public-private cloud interoperability could lock out other CSPs from tapping fully into this market. But other cloud players, such as Oracle and SAP, might develop tools with some hybrid capabilities, such as cross-segment databases.
Given that cloud portability isn’t easy, the hybrid component could be a key factor for some organizations when considering IaaS features. Microsoft shops will likely lean toward Azure, while Linux users will lean towards Google and AWS, the latter now also appealing to VMware shops.
Companies who run OpenStack might prefer Google because of its ties with Red Hat, while IBM’s role in in hybrid cloud might be the high levels of administrative automation it can offer with Watson as an AI tool.
Economic and strategic viability
As Hewlett Packard Enterprise and VMware have demonstrated, having a lot of money and a recognized brand in the IT world is not enough to win the public IaaS war. AWS, Google and Azure all started early in the cloud game and built a strong base of users.
The small CSPs face a struggle for survival. Rackspace is exiting the public cloud market to focus more on management, which opens up the question of whether smaller CSPs are sustainable. IBM and Oracle are somewhat niche players, but Watson and databases, respectively, might provide cover for them. Meanwhile, CSPs like Alibaba and Huawei maintain a strong market presence in China.